recommendation system

#TEXT_J_all_distanceo_cap_scale_0_1 <- sqrt(mulit_value_replace_distane^2+continues_euclideano_cap_0_1^2)



#weight_steamspy_tags_order_cosinedist<- cosinedist(weight_steamspy_tags_order)

#weight_steamspy_tags_order_jaccardc<-(1-jaccard(as.matrix(weight_steamspy_tags_order),10)[-1])[1:(length(cosine_dist_mulit_value))]



cosine_dist_mulit_value_PDN <- cosinedist(cbind(my_cosine,my_cosine_PDN))

mulit_value_replace_distane_PDN <- sqrt(cosine_dist_mulit_value_PDN^2+owners_jaccardo^2+eng_jaccardo^2+required_jaccardo^2+plat_jaccardo^2+(1.5*weight_steamspy_tags_order_cosinedist)^2)


#mulit_value_replace_distane_PDN <- sqrt(cosine_dist_mulit_value_PDN^2+owners_jaccardo^2+eng_jaccardo^2+required_jaccardo^2+plat_jaccardo^2+(1*cosine_dist_mulit_value)^2+(1*weight_steamspy_tags_order_cosinedist)^2)

TEXT_COSINE_all_distanceo_cap_scale_0_1 <- sqrt(mulit_value_replace_distane^2+continues_euclideano_cap_scale_0_1^2)


TEXT_COSINE_all_distanceo_cap_scale_0_1_PDN <- sqrt(mulit_value_replace_distane_PDN^2+continues_euclideano_cap_scale_0_1^2)

#parLapply(cl, all_distanceo_cap_scale_1_2, function(x,k=3) { cmdscale(x, k=3) })

cpu.cores <- detectCores()

cl <- makeCluster(cpu.cores)

#TEXT_COSINE_all_distanceo_cap_scale_0_1
#TEXT_J_all_distanceo_cap_scale_0_1
fitall343d <- cmdscale(TEXT_COSINE_all_distanceo_cap_scale_0_1_PDN,eig=TRUE, k=3)


x <- fitall343d$points[,1]
length(x)

## [1] 8727

y <- fitall343d$points[,2]
z <- fitall343d$points[,3]

#plot3d(x, y,z, xlab="Coordinate 1", ylab="Coordinate 2",
#  main="fite3d",col = colors)
#text3d(x=x, y=y, z=z,texts = row.names(train_datasetG_fifth), cex=.5)
#plot3d(x=x, y=y, z=z,xlab="PC1", ylab="PC2", zlab="PC3", size=5)
#text3d(x=x,y=y,z=z,texts=names, col=4)

#stopCluster(cl)

## Read the 8727 x 8727 data matrix successfully!
## OpenMP is working. 1 threads.
## Using no_dims = 3, perplexity = 50.000000, and theta = 0.500000
## Computing input similarities...
## Building tree...
## Done in 18.51 seconds (sparsity = 0.023570)!
## Learning embedding...
## Iteration 50: error is 89.806484 (50 iterations in 2.24 seconds)
## Iteration 100: error is 89.806268 (50 iterations in 2.71 seconds)
## Iteration 150: error is 89.153601 (50 iterations in 2.85 seconds)
## Iteration 200: error is 88.073210 (50 iterations in 2.11 seconds)
## Iteration 250: error is 87.936330 (50 iterations in 2.15 seconds)
## Iteration 300: error is 2.720462 (50 iterations in 1.86 seconds)
## Iteration 350: error is 2.382119 (50 iterations in 1.70 seconds)
## Iteration 400: error is 2.219792 (50 iterations in 1.71 seconds)
## Iteration 450: error is 2.120450 (50 iterations in 1.74 seconds)
## Iteration 500: error is 2.054506 (50 iterations in 1.69 seconds)
## Iteration 550: error is 2.006654 (50 iterations in 1.68 seconds)
## Iteration 600: error is 1.970315 (50 iterations in 1.67 seconds)
## Iteration 650: error is 1.941960 (50 iterations in 1.69 seconds)
## Iteration 700: error is 1.920616 (50 iterations in 1.67 seconds)
## Fitting performed in 27.47 seconds.

## [1] 8727

"mycex_posi <- c()
for (i in 1:length(PLOTER$categories1)) {
  if (PLOTER$positive_ratings[i] == summary(PLOTER$positive_ratings)[1]){
    mycex2[i] <- 30
  }else  if (PLOTER$positive_ratings[i] == summary((PLOTER$positive_ratings))[2]){
    mycex2[i] <- 20
  }else  if (PLOTER$positive_ratings[i] == summary((PLOTER$positive_ratings))[3]){
    mycex2[i] <- 10
  }else  {
    mycex2[i] <- 20
  }
  }

summary(PLOTER$positive_ratings)
table(mycex_posi)"

## [1] "mycex_posi <- c()\nfor (i in 1:length(PLOTER$categories1)) {\n  if (PLOTER$positive_ratings[i] == summary(PLOTER$positive_ratings)[1]){\n    mycex2[i] <- 30\n  }else  if (PLOTER$positive_ratings[i] == summary((PLOTER$positive_ratings))[2]){\n    mycex2[i] <- 20\n  }else  if (PLOTER$positive_ratings[i] == summary((PLOTER$positive_ratings))[3]){\n    mycex2[i] <- 10\n  }else  {\n    mycex2[i] <- 20\n  }\n  }\n\nsummary(PLOTER$positive_ratings)\ntable(mycex_posi)"

#summary(negative_ratings_encod_cap)
colors <- c()
for (i in 1:length(PLOTER$steamspy_tags)) {
  if (fifth_encod$steamspy_tags1[i] == "Action"){
    colors[i] <- "Action1"
  }else  if (fifth_encod$steamspy_tags2[i] == "Indie"){
    colors[i] <- "Indie2"
  }else  if (fifth_encod$steamspy_tags3[i] == "Indie"){
    colors[i] <- "Indie2"
  }else if (fifth_encod$steamspy_tags2[i] == "Action"){
     colors[i] <- "Action2"
  }else if (fifth_encod$steamspy_tags3[i] == "Action"){
     colors[i] <- "Action2"
  }else if (fifth_encod$steamspy_tags1[i] == "Indie"){
     colors[i] <- "Indie1"
  }else{
     colors[i] <- "none"
  }
  }
table(colors)/length(colors)

## colors
##    Action1    Action2     Indie1     Indie2       none 
## 0.11642030 0.10622207 0.05568924 0.17875559 0.54291280

#colors()

ply<- plot_ly( x=x, y=y, z=z, type = "scatter3d",text= PLOTER$name,marker = list(symbol = 'circle', sizemode = 'diameter'),color=colors ,colors=c("#191970",'#9400D3',"#E68AB8","white")
,size = PLOTER$positive_ratings, sizes = c(2,50),width = NULL)


#row.names(fifth_encod)
#c('green','black','white')
#c("#191970",'#9400D3',"#E68AB8","white")
?plot_ly

rating_log <- PLOTER$positive_ratings^1.5

plyt <- plot_ly(
  x = x1, 
  y = y1, 
  z = z1, 
  type = "scatter3d",
  text = row.names(fifth_encod),
  marker = list(symbol = 'arrow-down-open', sizemode = 'diameter'),
  color = colors,
  colorscale = list(
    c(0, colors[1]),
    c(0.25, colors[2]),
    c(0.75, colors[3]),
    c(1, colors[4])
  ),
  autocolorscale = FALSE,
  size = rating_log,
  sizes = c(2, 55),
  width = NULL,
  cauto = TRUE
)

#c('green','black','white')
#c("#191970",'#9400D3',"#E68AB8","white")

plot_ly Results

MDS

ply        

#,name=row.names(train_datasetG_fifth)
#sum(fitall343d$eig[1:2])/sum(fitall343d$eig)
#sum(fitall343d$eig[1:3])/sum(fitall343d$eig)

t-SNE

# 找出游戏 "Counter-Strike: Global Offensive" 的行号
game_idx <- which(row.names(fifth_encod) == "Yakuza: Like a Dragon")
# 将该游戏坐标设为原点
x0 <- x1[game_idx]
y0 <- y1[game_idx]
z0 <- z1[game_idx]

# 计算其他点相对于该游戏坐标的位置
x_relative <- x1 - x0
y_relative <- y1 - y0
z_relative <- z1 - z0

# 将 x, y, z 分别除以最大值，使所有坐标值都在 [0, 1] 之间
x_normalized <- x_relative / max(x_relative)
y_normalized <- y_relative / max(y_relative)
z_normalized <- z_relative / max(z_relative)

# 计算每个点到原点的距离，并按距离从小到大排序
dist <- sqrt(x_normalized^2 + y_normalized^2 + z_normalized^2)
idx <- order(dist)

# 根据距离排序的顺序重新排列 x, y, z 和其他相关数据
x_sorted <- x_normalized[idx]
y_sorted <- y_normalized[idx]
z_sorted <- z_normalized[idx]
color_sorted <- colors[idx]
rating_sorted <- PLOTER$positive_ratings[idx]

# 将距离最远的一些点的颜色设为白色，以模拟星空中的恒星
num_stars <- length(dist)
num_sky <- round(num_stars * 0.2)  # 设定 20% 的点为星空
color_sorted[(num_stars - num_sky + 1):num_stars] <- "white"

# 将评价数映射到 1-10 的范围内
rating_scaled <- scale(rating_sorted, center = min(rating_sorted), scale = max(rating_sorted) - min(rating_sorted)) * 9 + 1

# 绘制星空图
plot_ly(
  x = x_sorted, 
  y = y_sorted, 
  z = z_sorted, 
  type = "scatter3d",
  text = row.names(fifth_encod),
  marker = list(symbol = "circle", sizemode = "diameter"),
  color = color_sorted,
  size = rating_sorted,
  sizes = c(3, 35)
)

#size=(train_datasetG_fifth$positive_ratings_encod_cap)
plyt  

#,name=row.names(train_datasetG_fifth)

#one_hot(as.data.table(fifth_encod[,2:4]))

library(Matrix)
jaccard <- function(m,c) {
    ## common values:
    A = tcrossprod(m)
    ## indexes for non-zero common values
    im = which(A > 0, arr.ind=TRUE)
    ## counts for each row
    b = rowSums(m)

    ## only non-zero values of common
    Aim = A[im]

    ## Jacard formula: #common / (#i + #j - #common)
    J = sparseMatrix(
          i = im[,1],
          j = im[,2],
          x = (Aim+c) / (b[im[,1]] + b[im[,2]] - Aim+c),
          dims = dim(A)
    )

    return( J )
}

Matrix <- as.matrix(NORM)
sim <- Matrix / sqrt(rowSums(Matrix * Matrix))
sim <- sim %*% t(sim)

D_sim <- as.dist(1 - sim)

"NORM <- one_hot(as.data.table(fifth_encod[,2:4]))


tNORM <- t(NORM)


# NORM[1,],NORM[2,]
# tNORM[,1],tNORM[,2]

trydist <- matrix(rep(0,12881^2), nrow = nrow(NORM), ncol = nrow(NORM))
for (i in 1:10) {
  
  
   for(j in (i+1):10){
      
     trydist[i,j]  <- 1-jaccard(tNORM[,i],tNORM[,j],0.01)
     trydist[j,i] <- trydist[i,j] 
   }
}


jaccard(tNORM[,1],tNORM[,10])
Jaccard(tNORM[,1],tNORM[,20],0.01)

(1-jaccard(t(tNORM),0))[1:10]"

## [1] "NORM <- one_hot(as.data.table(fifth_encod[,2:4]))\n\n\ntNORM <- t(NORM)\n\n\n# NORM[1,],NORM[2,]\n# tNORM[,1],tNORM[,2]\n\ntrydist <- matrix(rep(0,12881^2), nrow = nrow(NORM), ncol = nrow(NORM))\nfor (i in 1:10) {\n  \n  \n   for(j in (i+1):10){\n      \n     trydist[i,j]  <- 1-jaccard(tNORM[,i],tNORM[,j],0.01)\n     trydist[j,i] <- trydist[i,j] \n   }\n}\n\n\njaccard(tNORM[,1],tNORM[,10])\nJaccard(tNORM[,1],tNORM[,20],0.01)\n\n(1-jaccard(t(tNORM),0))[1:10]"

#(1-jaccard(as.matrix(one_hot(as.data.table(fifth_encod[,1]))),0))[1:30]
#eng_jaccardo[1:30]


#plat_jaccardo[1:30]
#1-jaccard(as.matrix(one_hot(as.data.table(fifth_encod[,2:4]))),0)[-1][1:30] 
  
#sum(plat_jaccardo[1:3000]-(1-jaccard(as.matrix(one_hot(as.data.table(fifth_encod[,2:4]))),0)[-1][1:3000]))


#(1-jaccard(as.matrix(one_hot(as.data.table(fifth_encod[,2:4]))),0)[-1])



#jaccard(as.matrix(one_hot(as.data.table(fifth_encod[,5]))),5)[1:10]